Local Spillovers from Cash Transfer Programs: Food Price Increases and Nutrition Impacts on Non-beneficiary Children

Transcription

1 Local Spillovers from Cash Transfer Programs: Food Price Increases and Nutrition Impacts on Non-beneficiary Children Deon Filmer 1, Jed Friedman 2, Eeshani Kandpal 3, and Junko Onishi 4 Draft-- please do not cite; comments welcome October 18, 2016 Abstract: Cash transfer programs targeted to poor househ may increase prices, especially if local markets are not fully integrated into larger regional markets. Using data from the evaluation of a Philippine cash transfer program, we show that the prices of perishable protein rich foods exhibit moderate albeit sustained increases after program introduction. Likely as a result, anthropometric measures of child health, notably stunting rates, worsen among nonbeneficiary children. These effects are not short-run but persist up to 31 months after program introduction. Failing to consider the effect of such local price increases on non-beneficiaries wellbeing can overstate the impact of cash transfers. For very poor areas, where household targeting of cash transfers covers a majority of the househ, geographic targeting may avoid the consequences of local market price spillovers and consequently prevent nutritional and other long-run impacts at little additional cost. JEL Codes: H23; I38; O12; O15; O20 Keywords: Spillovers, cash transfers, price effects, food prices, Philippines 1 Development Research Group, World Bank, 1818 H Street NW, Washington DC 20433; dfilmer@worldbank.org 2 Development Research Group, World Bank, 1818 H Street NW, Washington DC 20433; jfriedman@worldbank.org 3 Development Research Group, World Bank, 1818 H Street NW, Washington DC 20433; ekandpal@worldbank.org 4 Social Protection and Labor Global Practice, World Bank, 1818 H Street NW, Washington DC 20433; jonishi@worldbank.org 1

2 Introduction Cash transfer programs provide cash to poor househ, whether conditional on the househ meeting some pre-specified behavioral criteria (CCTs) such as investing in their children s health or education or unconditional (UCTs). These programs reach between 750 million and 1 billion people (DFID, 2011), thus indicating widespread popularity among policy makers. While an extensive economic literature has considered both direct and indirect effects of cash transfer programs, the study of possible program spillovers has focused on the generally positive externalities that operate through social networks (Baird, Bohren, McIntosh, Ozler, 2011) or through informal insurance and credit markets (Angelucci and Di Giorgi, 2009). In contrast, this paper shows that even modest price increases resulting from cash transfers can result in significant detriments among non-beneficiary househ (in this case an increase in child stunting rates) that likely result in knock-on effects on human capital. The channel for this impact appears to be price increases of key protein-rich foods, which are particularly pronounced in very poor villages where a majority of househ are cash transfer program eligible. This paper thus recommends a reconsideration of targeting mechanisms when designing cash transfer programs in contexts where local markets aren t fully integrated with surrounding regions and a large proportion of househ in the village would receive program benefits under a household targeting rule. The Philippines has implemented a cash transfer conditioned on child health and education since This CCT, called the Pantawid Pamilyang Pilipino Program, or simply Pantawid, is targeted to individual househ based on a proxy means test for household income with eligibility cut-offs determined by province-specific poverty lines. Starting with an initial pre-pilot of 6000 househ, the Pantawid pilot reached 4.35 million househ by 2015 (DSWD, 2015). Indeed, Pantawid has reached more househ than many other national CCTs; by way of comparison, the Indonesian Program Keluarga Harapan (PKH) covered 1.5 million househ after 5 years (Nazara and Rahayu, 2013) and the fully-scaled up Mexican PROGRESA/Oportunidades program covered 5.8 million househ (World Bank, 2014). The government of the Philippines is currently considering yet another scale expansion of Pantawid. Given the large number of cash transfer programs and the number of househ covered by the various programs, it is not surprising that an extensive literature has studied both direct and indirect effects of cash transfers. This literature has been reviewed in Fiszbein and Schady (2009), Baird, Ferreira, Ozler and Woolcock (2013), Saavedra and Garcia (2012) and Hanlon, Barrientos and Hulme (2010). In general, evidence suggests that both UCTs and CCTs improve school enrollment and attendance as well as increase utilization of preventive health services, in addition to increases in total levels of household spending. Evidence also suggests that, holding income constant, beneficiary househ spend more on nutrient-rich foods than do untreated househ (Fiszbein and Schady, 2009). These results may be unsurprising because 2

3 enrollment, attendance and preventive health care are typically (a) normal goods whose demand increases as household resources increase, (b) behaviors on which CCTs are conditioned, and (c) both UCTs and CCTs are often accompanied by messaging underlining the importance of investments in human capital. However one might also expect there to be the possibility of spillovers to non-targeted outcomes as a result of the program. Potential spillovers operate at three different levels: within the household, within participating schools, hospitals and other facilities, and within affected local markets. Spillovers typically arise through two related channels: the information supplied by the program and/or changes in household spending, saving, and activity as a result of the income transfer. Within the household, Contreras and Maitra (2013) exploit program rules of a Colombian CCT to find significantly better health outcomes of non-targeted adults in treatment househ than in control househ. They present suggestive evidence that this improvement works through information. Ex-ante simulations based on Brazil s Bolsa Escola CCT also suggest that there may be changes in child work-for-pay, with full education subsidies generally reducing children s economic activities (Bourguignon, Ferreira and Leite, 2003). On the other hand de Hoop, Friedman, Kandpal, and Rosati (2015) show that the partial education subsidy offered by Pantawid was accompanied by an increase in child schooling and work-for-pay, while the larger subsidy, adequate for schooling costs, offered in PROGRESA/Oportunidades CCT led to an increase in child schooling but a decrease in children s economic activities. Ferreira, Filmer and Schady (2009) show conceptually and empirically using data from a Cambodian CCT that child-specific cash transfers generate positive income effects and negative displacement effects on the schooling of ineligible siblings, leading to an ambiguous overall effect on these siblings education. Within participating facilities, the literature provides robust evidence of peer effects-driven increases in schooling enrollment of non-targeted populations, at least in the case of PROGRESA/Oportunidades (Bobba and Gignoux, 2014; Bobonis and Finan, 2009; Lalive and Cattaneo, 2009). Of particular relevance for this study, Bobba and Gignoux (2014) use data from PROGRESA/Oportunidades to find strong evidence of positive externalities in potential beneficiaries school enrollment in areas with relatively few beneficiaries but less so in areas with a high proportion of beneficiary househ. They surmise that this effect likely arises from greater gains to information sharing in more sparsely treated areas relative to more densely treated areas. They note that this result underlines the importance of estimating program impact on sufficiently extended geographical areas. Within affected areas more broadly, evidence suggests a variety of positive externalities: From reductions in crime and political violence in the Philippines (Crost, Felter and Johnston, 2014) to an increase in saturation within social networks leading to higher test scores and lower HIV prevalence among the untreated in Malawi (Baird, Bohren, McIntosh and Ozler, 2011). Finally, 3

4 there is some evidence that Pantawid increased clientelism, which in turn improved political stability (Labonne, 2013). Examples of negative externalities include higher prices for health services resulting from an Indonesian CCT (Triyana, 2014) and production changes resulting in greater deforestation in communities with poor access to markets in Mexico (Alix-Garcia, McIntosh, Sims and Welch, 2013). Such local-area spillovers can also, in principle, affect non-beneficiaries consumption, such as through risk-sharing or effects on prices. However, evidence on such externalities is mixed: using data from Nicaraguan and Paraguayan CCTs, respectively, Macours, Schady and Vakis (2008) and Teixeira et al. (2011) do not find any evidence of externalities on non-beneficiaries. In contrast, studying the case of PROGRESA/Oportunidades, Angelucci and de Giorgi (2009) find that the CCT increased consumption by ineligibles through enhancing risk-sharing within household networks. A small literature has also looked explicitly at price externalities from CCTs: Cunha, De Giorgi and Jayachandran (2014) compare in-kind and cash transfers in an intervention in Mexico. They find that both types of transfers generate general equilibrium price effects for the local economy downward pressure on prices for the transferred good under the in-kind program, and positive pressure on price levels in the case of cash. On average, these price increases are modest and do not affect purchasing power. However, Cunha, Di Giorgi and Jayachandran report more sizeable price increases for remote communities, and indeed find that the cashversus-kind result is largely driven by price responses in these remote villages. Beegle, Galasso and Goldberg (2015) study another type of aid program a public works program (PWP) to find evidence that a large-scale PWP in Malawi significantly worsened food security for non-beneficiary househ in treated villages. Leveraging randomized treatment assignment of communities and househ within communities, the authors rule out pecuniary externalities as well as a crowding out of traditional risk sharing mechanisms, although they are unable to pinpoint the exact mechanism driving the negative spillover to untreated househ. Our study is therefore one of the few to identify an instance of local general equilibrium price effects from an aid program. To illustrate how price effects might arise from a cash transfer program, we present a conceptual model showing that increased cash in the village increases demand for normal goods, including many food goods. How supply responds to this demand increase, as well as the shape of the demand curve, determines the resulting local price level. The supply response to a demand increase can be constrained for a variety of reasons. For one, remote villages tend to suffer from high transport costs for imported goods, and the transport cost wedge may counteract the marginal gain in profit from importing more units to sell at a marginally higher price. For another, if local production markets are either oligopolistic (perhaps due to a fixed 4

5 cost of entry) or competitive with upward sloping MC then price increases will also likely follow a positive shift in demand. It turns out in this setting that price increases are observed for perishable protein-rich foods fresh eggs, fresh fish, and other goods with relatively high importation costs from outside the local market but not for storable food goods such as rice or packaged food that are lower in protein but higher in other nutrients. This increase in relative prices results in a drop in real income for non-beneficiaries and a substitution away from protein-rich foods. The relative price changes and shifts in consumption are not large, but they are apparently enough to be expressed through lower child health as the growth of young children are particularly vulnerable to deficiencies in the intake of protein (Puentes et al., 2014). The next section discusses the conditions for local price changes arising from cash transfers in more detail, while the subsequent section describes the Pantawid program and the data used in this analysis. The presentation of results follows and then the paper concludes with a discussion of the findings and implications for the targeting of social programs. Possible price responses to cash-transfer induced demand increases Although direct and indirect effects of cash transfers have been extensively studied, our understanding of the general equilibrium effects, and the economic mechanisms through which the cash and the conditions affect entire communities, remains incomplete. In contrast to previous literature that finds either no or positive spillovers on non-beneficiaries consumption, we document a case of negative spillovers on the non-targeted population s consumption and nutrition. The relative price change of key protein-rich food goods appears to be the main driver of this outcome. How might this relative price change arise? For one, the income gain from cash transfers increases demand for normal goods and if supply does not fully respond then prices will rise. In addition, conditional cash transfer programs such as Pantawid also broadcast messaging on nutritious foods that may additionally shift relative demand for these goods in particular. As potential price change is related to the magnitude of demand increase, one relevant program aspect that determines the magnitude of price change is what we call program saturation, namely the proportion of the population that benefits from the program. The higher the degree of program saturation, the greater the increase in aggregate demand and hence the greater potential for relative price change. Thus if we are able to observe resultant price increases at all, we should observe them in poorer villages that have a high proportion of program beneficiaries. Besides the magnitude of the increase in aggregate demand, the price response will also depend on the structure of the market supplying the good. If the markets that supply these 5

6 goods are perfectly competitive and have access to a broad regional or national (or international) production base, then suppliers should be able to meet any demand increase with relative ease. In this case, the only way price would rise is if the increase in production to meet the new demand raises the marginal cost of production. If the Philippines operated as one integrated market for the good in question, the total increase in aggregate demand due to Pantawid would likely have to be very large to raise the marginal cost of production to a noticeable degree (while widespread, program beneficiaries are only a small percentage of the total population). So a price rise in this case is unlikely. However, the assumption of perfectly competitive food suppliers with access to a broad production base may not be fully applicable in the study villages, at least for all types of food goods. Poor villages can suffer from high transport costs and often have many basic goods locally sourced. This is especially true for perishable goods such as fresh fish or eggs that would need special technologies for transport and storage. For these goods, which we term nontradable (i.e. across villages) for ease of exposition, the elasticity of supply may be low, especially with respect to relatively small changes in aggregate demand. As a result, price change induced by cash transfer programs can be heterogeneous not only as a result of shifts in relative demand but also due to the degree of integration of the local goods market. Thus another factor determining the price response, besides program saturation, is the tradable nature of the specific good in question. Our study will explore price responses along both these dimensions by contrasting low with high program saturated villages as well as contrasting tradable goods, such as rice and packaged food, with non-tradables such as fresh eggs and fresh fish. Now with a refined focus on local market price changes for non-tradable goods, a rise in demand may or may not result in a price rise depending on the nature of the local market. A competitive local market with a flat MC curve will see no change in price and one with a rising MC curve will see a rise in price even if the local market is able to fully satisfy the new demand with an increase in production. Perhaps a more realistic model of the local market involves imperfect competition as there are relatively few producers of any local good, barriers to entry in terms of fixed costs such as a fishing boat or the cost of livestock, and perhaps even the possibility for collusion. Under imperfect competition, price increases from a rise in demand will occur even if the marginal cost of production is constant. The price response to a demand shift of a given magnitude under imperfect competition depends on the shape of the demand curve, the number of producers in the market, and the fixed cost of entry for potential competitors. In general, the greater the number of firms operating, the smaller the price increase from a demand shift. Unfortunately we do not observe the number of local producers for perishable food goods, the fixed cost of entry, or the price threshold that may entice traders to import the good into the village. Thus we will not be able 6

7 to say anything directly about the structure of the local market. Instead we will investigate the observed degree of price change and how the magnitude of change is or is not correlated with the factors that are likely to affect a relative price rise in the case of non-fully integrated markets, i.e. the degree of program saturation and the tradability of the good. Cunha et al. (2014) conduct a similar inquiry in rural Mexico and find that cash transfers led to a positive but negligible increase in price. 5 However that is estimated over the entire sample as described above, price changes can be greater in areas where supply responses are constrained. Indeed the authors find that the price change as a result of the cash program is more pronounced in remote villages, where prices increase for basic foods goods on the order of 6%. Geographic isolation is used in this study as a proxy for how closed an economy is (low elasticity of supply) or for how uncompetitive the market is (low number of producers). In extensions of the analysis we will also use remoteness indicators as a supplementary measure for the integration, or lack thereof, of the local market with wider regional ones. Any observed price increase may not be too large as otherwise arbitrage opportunities would arise that would compensate for the cost of importation of goods into the village. In this sense the transport cost of goods drives a wedge between local and national prices that may persist over time even if the local price rises from a previous equilibrium price. Cunha et al. find that price effects persist for up to 22 months, suggesting that local supply responses do not (fully) counteract the demand increase from the income transfer. Our results are measured after 31 months of program exposure, and any evidence of a price rise would suggest the same. The Pantawid program and evaluation design The Philippines Conditional Cash Transfer (CCT) Program, Pantawid Pamilyang Pilipino Program, provides cash transfers to poor househ, conditional upon investments in child education and health as well as use of maternal health services. Eligible poor househ are identified by the National Household Targeting System for Poverty Reduction (NHTS-PR) based on a poverty targeting mechanism using a proxy means test, which estimates per capita household income on the basis of observable easily answered information such as household size and the physical conditions of the dwelling. Househ with estimated per capita income below the poverty line are classified as poor. From this database of poor househ, Pantawid identifies and selects eligible househ who have children 0-14 years of age and/or a pregnant woman at the time of the assessment. Poor and eligible househ receive a combination of health grants and education grants every two months ranging from PhP 500 to 5 They also look at in-kind transfers, which also increase supply as well as transfer income, and found a 4% decline in prices among the in-kind villages. 7

8 PhP 1,400 (approximately 11 USD to 32 USD) per household per month, depending on the number of eligible children in the household. Besides family size, the exact transfer amount is also determined by the compliance behavior of the household with respect to two types of grants: Health Grants. The health grant is aimed at promoting healthy practices, improving the nutritional status of young children, and increasing the use of health services. Poor househ with children 0-14 years old and/or pregnant women receive a lump sum amount of PhP 500 (about US$ 11) per household per month. Househ must fulfill the following conditions for the health transfer: (i) all children under the age of five follow the Department of Health (DOH) protocol by visiting the health center or rural health unit regularly; (ii) pregnant women attend the health center or rural health unit according to DOH protocol; (iii) all school-aged children (6-14 years old) comply with the de-worming protocol at schools; and (iv) for househ with children 0-14 years old, the household grantee (mother) and/or spouse shall attend Family Development Sessions at least once a month. One major topic of these sessions is family nutrition, which encourages the consumption of fresh protein-rich foods and deemphasizes the consumption of packaged foods. Education Grants. The education grant is aimed at improving school attendance of children 6-14 years old living in poor househ in selected areas 6. The education transfer is PhP 300 (about US$ 6.50) per child per month (for a period of 10 months/year), for up to a maximum of three children in the household. Beneficiary househ receive the education transfer for each child as long as they are enrolled in primary or secondary school and attend 85 percent of the school days every month. Since its program launch in 2008, Pantawid has scaled up rapidly and has become the cornerstone of the Government s social protection strategy. By December 2014, the program had approximately 4.45 million active beneficiary househ. As a requirement of initial donor financing, an evaluation was designed from the early stages of the Pantawid program and viewed as an integral part of the program by implementers and policymakers. The first round of the impact evaluation was intended to represent the first implementation phase of the program, as the program s scale-up plan was not yet in place at the time of study design. This first phase covered some of the poorest areas of the country and the study purposively selected eight municipalities to be included in this phase for the evaluation. The Household Assessment Form (HAF) to estimate proxy means scores for beneficiary selection was fielded in these eight municipalities between October 2008 and 6 The education grants were later extended after the study period to cover students up to 18 years old. 8

9 January This was followed by the implementation of Pantawid in the treated villages, with the first payment of cash grants commencing in April The evaluation is a randomized control trial, stratified at the municipal level, with randomization at the village (also known as barangay in Filipino) level. A total of 130 villages were equally likely to be randomly selected to treatment or control status in the eight municipalities 7 selected for the impact evaluation 8. A follow-up survey was conducted in October and November 2011, allowing a program exposure period of 30 to 31 months. Data A total of 3,742 househ were surveyed from the eight study municipalities during the follow-up survey. With an eye toward investigating potential spillovers on non-beneficiary househ, the entire study population was divided into four categories using the National Household Targeting Survey database as follows 9 : 1,418 Category 1 househ that were designated poor househ (below the PMT score) with children aged 0-14 or a pregnant mother at the time of the household assessment (the eligible group for Pantawid); 1,137 Category 2 househ designated non-poor househ (above the PMT score) with children aged 0-14 or a pregnant mother; 556 Category 3 househ that were poor househ yet without children aged 0-14 or a pregnant mother; and 631 Category 4 househ that were non-poor but without children aged 0-14 or a pregnant mother. Direct effects of the Pantawid program are estimated by comparing Category 1 househ in treated and control villages. Possible spillover effects outside the beneficiary household but 7 The study sample for the impact evaluation was selected in three stages. First, provinces in which the program had not yet been introduced in some of the eligible municipalities as of October 2008 were selected. Out of the 11 provinces available, 3 provinces were excluded due to security concerns. From the remaining 8 provinces, 4 provinces were chosen to span all three macro areas of the country (North, Visayas, and Mindanao). Second, among the selected four provinces, municipalities were randomly chosen to represent the average poverty level of areas covered by the program. Two municipalities each were selected for the study in the provinces of Lanao Del Norte, Mountain Province, Negros Oriental, and Occidental Mindoro. The set of provinces and municipalities for the RCT was selected jointly by DSWD and the World Bank, and barangay randomization was conducted in October The eligible household in the control villages started receiving the program benefits immediately after the survey was completed. 9 The sample was designed to identify spillover effects to non-beneficiary target groups, as well as to run the RD analysis on the data from RCT sample areas. 9

10 within the village can be explored with impact estimates of Category 2, 3, and 4 househ. Because of the focus on child nutrition, this paper here will present basic program impacts among Category 1 househ and then explore possible spillover effects among Category 2 househ. In terms of program coverage, the impact evaluation survey and program Management Information System (MIS) database yielded slightly different estimates. Although all of the 1,418 househ in Category 1 were eligible to become Pantawid beneficiaries in 2008, only those in treated villages were offered the program in 2009 by design. Among the 704 Category 1 househ sampled in the Pantawid villages, 85 percent (581) reported being beneficiaries of the program, while 1 percent (7) in the control villages also reported being beneficiaries. According to the program Management Information System (MIS) database, however, the control villages did not have any beneficiary househ, and 91 percent (647) of the 704 sampled Category 1 househ in the Pantawid villages were considered beneficiaries of the program. 10 Small numbers of househ among Categories 2, 3, and 4 (5 percent, 5 percent, and 10 percent, respectively) reported being Pantawid beneficiaries, even though none of these househ were program beneficiaries according to the program MIS database. The survey data include complete height-for-age data on 172 non-beneficiary children 6-36 months of age in treated areas and 151 non-beneficiary children of the same age range in control areas. Weight-for-age data were collected for month old non-beneficiary children in treated areas and 156 non-beneficiary children in control areas. Anthropometric z- scores were calculated based on the WHO (2006) growth standard. Scores of more than 6 standard deviations above or below the reference mean were dropped from the sample (Rutstein, 2006). This trimming resulted in 14 of the 172 treated children and 6 of the 151 control children being dropped from the height-for-age regressions, and 2 of the 177 treated and none of the 156 control children being dropped from the weight-for-age regressions. Annex 1 explores alternative cutoffs for trimming the data, and shows that the results are robust to the data trimming. 10 The lower percentage of sampled househ in Pantawid villages that reported being program beneficiaries may be explained in part by the fact that program participation is voluntary. Some househ identified as potential beneficiaries may have waived their right to the program. Another possibility is that through the community validation process of NHTS-PR, these househ may have been taken off the list of poor househ. It is also possible that a potential beneficiary household was unaware of the community assembly where attendance is required for potential beneficiaries to sign up for the program and confirm their basic household information collected for the PMT. Although very small in number, it is more difficult to explain why non-beneficiary househ according to the program MIS reported themselves to be Pantawid beneficiaries in the survey. There is no official way for a household that was not identified as poor by the NTHS-PR to be registered as a Pantawid beneficiary. It is possible that the respondent was thinking of some other program that they received rather than Pantawid. 10

11 In addition to survey data, this paper uses a time series of item-specific food prices collected by the Bureau of Agricultural Statistics. This data was collected monthly from and reported at the provincial level. In each province, price enumerators visit six markets four rural and two urban however the location of these markets are unknown to researchers. As this price data serves as an input to the Consumer Price Index of the Philippines, rigorous field controls are used to ensure the quality comparability of goods assessed (Philippine Statistics Office, 2015). Experimental Evaluation of Pantawid: Balance and Results on Beneficiaries Since program saturation is the key mediator of the cash transfer s impact on prices of nontradables, we examine the expansion of Pantawid coverage by looking at the provincial data on the annual change in program saturation of Pantawid from 2006 to Indeed, Figure 1 illustrates the rapid expansion of Pantawid in this time period. This high rate of increase in the program suggests that within-village saturation rates may also have been high, generating conditions that may have resulted in prices increases, as discussed above. <Figure 1 here> As the municipalities selected at this stage of the program were among the poorest in the Philippines, it is no surprise that in many of the study barangays, a high proportion of the total population was eligible to receive program benefits. Figure 2 presents a histogram of this village level proportion of eligible househ from among all househ with children in the village we term this proportion saturation as described above for both treatment and control barangays. While there is a good degree of dispersion in this saturation measure, some villages have up to 90% of the household population eligible to receive benefits and in the typical village the majority of househ are eligible. The median barangay saturation level is 65% and the mean 62% for the entire study sample. <Figure 2 here> The HAF that determined the household proxy means score constitutes the baseline date for the Pantawid evaluation. This information, relatively limited in scope, is primarily used to assess characteristic balance across treatment and control villages for the socio-demographic and economic information collected. Appendix Tables 1 & 2 explore baseline balance for Category 1 and 2 househ respectively. These tables suggest that, overall, these categories were balanced. For category 1, only one comparison out of 28 is imbalanced treated Category 1 household heads are three percentage points less likely to have completed some high school. However, this lack of balance is unlikely to be of import for the analysis presented here, as we primarily rely on comparisons within Category 2. For Category 2, balance is almost as 11

12 comprehensive: only two out of 28 comparisons between treated and control areas are significantly different. While overall wealth, as measured by the logged PMT score is perfectly balanced between treated and control areas, Category 2 househ in treated areas are less likely to own video recorders or motorcycles. <Appendix tables 1 & 2 here> Since saturation of the village is an important mediator to assess spillover impact, Appendix Tables 3 & 4 also present baseline characteristic balance in Category 2 househ for those above and below the median saturation level. These tables highlight the overall balance of the experiment. For Category 2 househ in above median saturation areas, 25 out of 28 comparisons, including the mean proxy of wealth, household head s education, and children s school attendance are completely balanced. Treated househ are slightly smaller, and slightly less likely to have strong roof materials or a telephone than househ in control areas. Similarly, in below median saturated areas, househ are balanced along 26 of 28 dimensions. Treated househ are slightly less likely to own their house or a video recorder, although here again the aggregate wealth proxy is balanced. <Appendix tables 3 & 4 here> Fielded in October and November of 2011, the follow-up survey was directed both at househ and community respondents and covered an extensive range of socio-economic information including child anthropometric measures for ages 6-60 months (which were not assessed at baseline). Therefore, using the baseline characteristics available from the listing data, it appears that the experiment was well balanced. Pantawid incentivized the health and education related behavior around children in beneficiary househ. Table 1 presents some of the main impacts among beneficiaries of the program on outcomes related to these targets. Similar to findings from other CCT programs, the enrollment and attendance of children in the targeted age ranges improves on the order of 4 percentage points in terms of enrollment and 2-3 percentage points in terms of attendance, depending on the age group analyzed. These improvements were identified despite an already high level of enrollment and attendance in the control communities. A range of nutrition indicators was investigated, as reducing childhood malnutrition is one of the main goals of Pantawid. The considered age group for these indicators in Table 1 is children 6-36 months old as these children transit a critical developmental period for physical growth. Children in this age range also are likely to have lived most or all of their lives exposed to the program. While there is no precisely estimated impact on the mean height-for-age score or other anthropometric measures (the point estimates suggest an improvement of 0.3 standard deviations in the z-score and reduction in stunting likelihood of 2 percentage points), the 12

13 program lowered the rate of severe stunting 11 among poor children 6-36 months old by 9.3 percentage points. Stunting is a measure of chronic malnutrition, reflecting extended periods of inadequate food intake and/or chronic infection. No program impacts were found on other measures of severe or acute malnutrition such as wasting 12 or severe wasting. 13 For the average beneficiary child there may not have been a noticeable improvement in nutrition status but for the most disadvantaged there was a marked improvement. <Table 1. Program impacts on beneficiaries education and nutrition - here> One of the ways in which the cash transfer may have resulted in improvements in child anthropometry is if beneficiary househ seek to consume more of the goods associated with increases in child height-for-age. We look for evidence of this in two ways, first with respect to reported spending patterns of various food goods and then with regard to the reported food intake of young children. The second column of Table 2 reports the program impacts on household food budget share. For a select number of individual food goods, we can also investigate the household reported item consumption for young children under 60 months of age. These results are presented in columns 3-6 of Table 2 and are based on recall over the week before survey. Among beneficiary househ, the cash transfer should increase available resources for spending. Indeed, we find that the total foods share of the household budget actually declines a modest degree (by 2.9 percentage points), indicating that househ are moving along the food Engle curve as predicted after a gain in income. Among beneficiary children in Pantawid villages, there was a 8.2 percentage point increase in parents feeding their children (0-5 year who are fed solid foods) eggs, as well as some indication of higher meat and fish frequency (although not precisely estimated) during the previous week compared to children in nonprogram villages. <Table 2. Household expenditure impacts for beneficiaries and non-beneficiaries> Price, expenditure, and consumption effects The first step in the proposed causal chain leading to nutrition deficits as a result of program pecuniary spillovers to non-beneficiaries is the presence of higher food prices. We investigate this with two sources of price information. First, we analyze price changes with official itemspecific price series data. This data is reported at the provincial level for all 81 provinces in the Philippines on a monthly basis and covers a period starting in 2006, two years before the 11 Measured as height-for-age <-3SD applying the WHO Child Growth Standard ( accessed March 9, Weight-for-age <-2SD applying the WHO Child Growth Standard 13 Weight-for-age <-3SD applying the WHO Child Growth Standard 13

14 introduction of Pantawid, and extending through the entire scale-up period that concluded in As the price data is reported only at the provincial level, and not at an administrative level below that, we relate changes in province specific prices to changes in a provincial level saturation measure of Pantawid exposure. This measure is calculated as the number of beneficiary househ reported by the provincial-level office of the Department of Social Welfare for that year divided by census bureau estimates for the total number of househ in the province. We present prices for three perishable goods fresh eggs, fish, and chicken and three tradable goods rice, snacks and sugar. The average annual provincial price for each good, Pipy, is regressed according to the following specification: P ipy = γ 0 + γ 1 S py + F p + F y + ε ipy where S is the province-year specific saturation measure and i, p, and y index good, province, and year. The specification also includes province and year fixed effects, Fp and Fy respectively. The coefficient of interest, γ 1, captures the good-specific price deviation from its provincial mean level, net of common year effects, as a function of the mean-differenced changes in provincial exposure. The three non-tradable goods all exhibit price increases correlated with changes in program saturation at the provincial level (Table 3). The maximum saturation level in the province level data is.40, suggesting that provincial prices for eggs can rise as much as 7.7% (0.192*0.40) as a result of price spillovers from the Pantawid program. Maximum price increases are on the order of 5-6% for fresh fish and chicken. In contrast, the three tradable goods show no significant price co-variation with program saturation as would be predicted if, even in poorer high-saturated villages, traders can access larger more integrated markets to satisfy a rise in food demand as a result of the cash transfer. The magnitude of the non-tradable goods price increase are not large, although it s unlikely that substantially higher increases would be able to sustained as larger increases may lead to arbitrage opportunities. <Table 3. Impact on provincial level prices, here> Besides province-level price changes, we also examine changes in the unit values of individual food goods reported by the survey respondents. This information was recorded only for three individual goods of standardized quality eggs, rice, and sugar. The first panel of Table 4 explores how relative price levels vary at the time of survey between program and control villages. None of the prices are significantly different, and the point estimates of the rice and sugar are close to zero as well. While also not precisely estimated, the point estimate for egg price stands at almost 2% higher, indicating some divergence in relative price difference between the storable goods such as rice and sugar and the perishable good, eggs. 14

15 Relative price differences emerge much more clearly when the program indicator is interacted with the binary measure of high saturation villages. This interaction effect indicates a relative price increase of 0.36 pesos per egg (0.06*6.015) in saturated villages, a rise of approximately 6%. The price changes for the tradable goods rice and sugar are close to zero in magnitude and not precisely estimated. As eggs are the most perishable good in this three good comparison, the price divergences are consistent with the predictions discussed above. We observe a price rise in program villages, but only for the non-tradable good, in highly saturated villages. It s an open question whether uncompensated price changes of these magnitudes are large enough to shift demand choices of the non-beneficiary househ, especially those with children. We return to the survey data to investigate this next question. <Table 4. Impact on village level prices here> Results on non-beneficiaries So far, the analysis has demonstrated that the program improved health and education outcomes of children from beneficiary househ concomitantly raising children s consumption of protein-rich foods. The analysis has also identified a rise in the price of selected non-tradable goods, but not in more easily traded goods, over the course of Pantawid introduction that is correlated with program saturation measures. We identify this general price change pattern with two independent sources of price data. However, such an increase in food prices may also have affected the consumption of these foods by non-beneficiary househ. Table 5 thus presents results that parallel Table 2, but this time contrasting Category 2 in treated and control villages, and reports the program impacts on household food budget share as well as whether the household reported feeding eggs, meat, and fish to children 6-60 months old. For non-beneficiary househ in treated villages, food expenditure as a share of household budget significantly increased by 3.6%, which suggests a decline in real income through the rising local prices of perishable foods, and perhaps a substitution away from dairy and eggs and towards cereals. It s difficult to infer too much from the spending data, although the change in patterns between treatment and control villages is consistent with a rise in demand for protein rich foods (as well as greater spending on other child goods) among beneficiary househ, and perhaps a substitution away from protein rich foods for nonbeneficiaries. <Table 5. Household expenditure and children s food intake impacts for non-beneficiaries> The program appears to also have had impacts on feeding practices, although not in all aspects. With the provision of cash coupled with parenting education provided during the program s Family Development Sessions, the program was expected to have some impacts on parenting 15

16 practices, including feeding practices. Indeed, food intake among non-beneficiary children doesn t change nearly as much as a result of the program the point estimates for the intake of eggs and vegetables are positive although not precisely estimated, suggesting little change in impact. However, as the price changes were seen in highly saturated villages, the food consumption intake of non-beneficiary househ in those villages may be appreciably different. The bottom panel of Table 5 thus explores food intake in these villages through fully interacting program exposure with an indicator for above median saturation. Note that in this decomposition, the incidence and quantity of egg consumption among non-beneficiary househ is higher in Pantawid villages. This may be due to informational spillovers of the program itself and the messaging around nutritious food can also be absorbed by the nontargeted househ. Egg consumption also appears greater in highly saturated villages in general. This can be due to various unobserved differences at the village level since high saturation villages are poorer on average and may differ in other key characteristics that determine demand patterns. The interaction term, however, is strongly negative. The lower incidence of egg consumption for these children when compared with children in highly saturated control villages (or compared with children in low saturated but treated villages) is immediately apparent. The same h for the number of eggs consumed, meat and fish, although these the effects on these three variables are not precisely estimated. The question then arises: did the price increases and concomitant decreases in non-beneficiary children s consumption of protein-rich foods in highly saturated villages affect their anthropometry or educational outcomes? Direct and indirect anthropometric impacts The top panel of Table 6 presents the same schooling and nutrition measures as presented in Table 1 but now contrasts Category 2 househ in Pantawid and non-pantawid villages. Regarding school related outcomes, non-beneficiary househ exhibit little change in the enrollment or attendance of school-age children. These levels are already near universal and substantially higher than the enrollment or attendance of beneficiary children residing in the poorer targeted househ in the village. As these children are not enrolled in the program it is perhaps little surprising that schooling-related indicators do not change after program introduction, although it does suggest that there are few schooling specific spillovers in terms of higher fees or increased crowding that may deter the attendance of non-beneficiary children. It is a different story for child anthropometric measures, here presented for non-beneficiary children 6-36 months of age in the second panel of Table 2. Children in non-beneficiary househ are substantially shorter if they reside in program barangays z-scores shorter than their counter-parts in barangays without the program. They are also significantly more likely to be stunted. The stunting rate is estimated at 32% in control barangays compared with 16

17 43% in barangays with the Pantawid program. While the point estimate for weight-for-age is also negative, it is not precisely estimated, suggesting particularly pronounced effects among longer-term nutritional measures such as child height. If increases in the prices of protein-rich foods, and the concomitant decrease in non-beneficiary children s consumption of these foods are associated with the worsening nutritional outcomes, then we would also expect to see the strongest nutritional effects in the villages where the price increases are the biggest: villages with the highest rates of program saturation. The second panel of Table 6 presents the impacts of living in an above-median saturated Pantawid village on children s schooling and nutrition, while the bottom panel presents the impacts of living in a village with a Pantawid saturation rate in the fourth quartile. Indeed, we find that weight-for-age is significantly lower and the likelihood of being underweight significantly higher in program villages that have high rates of saturation. Average height-for-age is lower and stunting rates higher in highly saturated villages, but the coefficients are precisely estimated for only the villages in the fourth quartile. On the other hand, average weight-for-age is significantly lower and the underweight prevalence significantly higher in above median saturated treated villages as well as treated villages in the top quartile of saturation. <Table 6. Impacts on non- beneficiaries, education and anthro here> While the study randomization of program villages resulted in a highly balanced sample across the characteristics assessed at baseline, child growth indicators were not measured. However as child growth is particularly sensitive to nutritional and health conditions in the first 1000 days of life (Hoddinott et al., 2013), we can investigate the age-patterns of child height differences among those who lived much of the first 1000 days under the program compared with somewhat older children born and partially reared before program onset. If the nutritional impacts on non-beneficiary children can be attributed to program presence and not other unobserved factors, then we would not expect to see the same impact among older children. Figure 2 depicts the proportion of children stunted between treated and control barangays by three age ranges. The stunting prevalence for children months, and hence only partially exposed to the program at critical ages for growth, is virtually identical. This is not the same among younger children where the stunting rate is substantially higher in treated barangays both for 6-24 months old, and <Figure 2 Anthro impacts by age, here> These age differences suggested by Figure 2 are apparent in the impact regressions in Table 7 that now investigate nutrition impacts pooled among 6-60 with the program exposure indicators interacted with the younger age categories of 6-23 months and months. HfA z-scores are significantly lower for the 6-23, i.e. those children who 17

18 have been exposed to the program for the entirety of their lives (and in-utero as well), on the order of.70 standard deviations. Stunting rates are also higher (15 percentage points) but the impact is not as precisely estimated. Further, the impact on weight-related nutrition measures, which capture shorter-run measures of health status, also emerge for this age group. Younger non-beneficiary children are significantly more likely to be underweight, on the order of 20 percentage points. For nonbeneficiary children months, the point estimates of impact also suggest a worsening of nutritional status but to a lesser degree there is no difference in wasting for example and the difference not as precisely estimated. Taken altogether, if children in non-beneficiary househ suffer growth deficits as a result of the program then we would expect to see a divergence in growth only for those children under an age cut-off when they are most vulnerable to a nutritional deficit. We see this for children under 36 months, and especially for those 6-24 months old at the time of survey. <Table 7. Anthro impacts on non-beneficiaries by age, here> At this point in the analysis, we have identified nutrition gains among beneficiary children and deficits among non-beneficiary children as a result of program exposure. Price increases of key foods are also correlated with program exposure, suggesting a key channel for program spillovers to non-beneficiary children. Observed spending patterns and reported food intake of young children are also somewhat consistent with spillovers operating through this channel. Other channels We present the evidence above to support the hypothesis that the high saturation of Pantawid increased the prices of certain non-tradable foods that are important for the production of child height, leading to increased stunting among non-beneficiary children in these highly saturated areas. In this section, we investigate other competing hypotheses and whether other household behaviors support our hypothesis. For instance, it possible that the observed increases in child malnutrition among nonbeneficiary househ may also have been caused by a lack of balance at baseline rather than by Pantawid s impact on prices. As shown in Appendix Table 3, non-beneficiary househ in above median saturated treatment areas were significantly smaller (by 0.41 people, particularly adults) than control non-beneficiary househ in above median saturated control areas, which in turn may affect the number of caretakers available for young children and thus household responsiveness to child illness or the available household resources that can be devoted to children. In Table 8, we use baseline data on household composition for a differences-in-differences approach to investigate the potential differences in household composition across Pantawid and non-pantawid villages among non-beneficiary househ 18

19 with young children. We find no significant differences in household composition or in household dependency ratio, whether overall or with respect to female or male caregivers, in treated areas 31 months after rollout. The triple difference with program saturation, presented in the lower panel of Table 8, also does not suggest differences in household composition, although the average treated household has 0.36 more people. If the lack of baseline balance in household size between non-beneficiaries in saturated villages had been driving difference the increases in child malnutrition, we would have expected to see the triple difference terms to be precisely estimated. <Table 8. Impact on household composition for non-beneficiaries, here> On the other hand, if real incomes decline for non-beneficiary househ as a result of price changes, as predicted by the theoretical framework, adult household members may respond by increasing their labor supply to compensate for the fall in real income, thus reducing the availability of adult caregivers. Table 9 looks at the labor force participation, work-for-pay and full-time work (greater than 40 hours per week) for adult men and women in non-beneficiary househ. Overall there is little change in labor force participation or hours worked for either men or women. The second panel, which contains the fully interacted model between the program indicator and village saturation, also shows little evidence of change in male or female labor force participation in above median saturated non-beneficiary househ. <Table 9. Impact on adult LFP among non-beneficiaries> If our hypothesis is correct, then the observed spillovers on non-beneficiary househ work through the specific mechanism of increased food prices that affect child growth. While possible confounders may be the lack of caregivers, examined above, and possibly quality of care, examined further below, we should not necessarily see any impacts on non-beneficiary children s education or participation in the labor force. To determine if this is in fact the case, Table 10 looks at the education and child labor impacts of Pantawid on non-beneficiary children. The top panel of Table 10 shows the basic impact of the program, while the bottom panel interacts treatment with saturation. None of the estimated effects of Pantawid are significant, suggesting that the spillovers caused by Pantawid are indeed quite specific. <Table 10. Impact on children s education and LFP among non-beneficiaries> Finally, we note that child height is determined in early life not only by nutrition but also exposure to infections and other pathogens. Thus another potential channel is through program impacts on the access to early life health services and the quality of those services. This is especially important to investigate since maternal and child care is directly incentivized by Pantawid. To the extent that the formal health care sector is able to improve child health, a degradation in either the access to or the quality of health services can also in principle 19

20 contribute to increased stunting. The Pantawid program may degrade the access to, or quality of, care through the crowding-out of available services as a result of an increase in service utilization by beneficiary househ. This crowding out mechanism can result either in increased prices for care, through impacts on the quality of available services, or both. In neighboring Indonesia, Triyana (2014) finds that a CCT conditioned on safe-delivery practices results in a 10% increase in fees charged by mid-wives. 14 Table 11 investigates program impacts on a range of health care seeking behavior relevant for young children for beneficiary and non-beneficiary househ, as intended. The first panel of Table 11 indicates that, among beneficiaries, Pantawid has increased the use of maternal and child health services such as antenatal care, postnatal care, skilled birth attendance, growth monitoring, and general treatment seeking (the first three measures are based children on 6-36 months old since they refer to care around the birth, while the last two measure of more general care seeking are based on children 6-60 months). The Pantawid program has been very successful in getting beneficiary househ to increase utilization rates, albeit these househ start from a very low base. The second panel of Table 11 investigates the impact of Pantawid on care-seeking by nonbeneficiary househ. Perhaps worryingly, non-beneficiaries in treated areas had significantly fewer ANC visits skilled birth attendance. However the changes in program utilization for these indicators are unrelated to the program saturation. This suggests that service accessibility around delivery is not related to the growth deficits observed in non-beneficiary children. And in general, the increased utilization from non-beneficiaries is not related to a change in utilization among children of non-beneficiary househ. We are unable to look at the prices of such utilization or, perhaps more importantly, the quality of services delivered. However the initial results suggest that changes in health care accessibility are not a major contributor to increased stunting. <Table 11. Impact on health seeking behavior for beneficiaries and non-beneficiaries, here> Conclusions and discussion This study investigates the impact of the Philippines flagship anti-poverty program on a range of outcomes for beneficiaries and non-beneficiaries. We show that the direct effects of the program on recipients were consistent with what has been found elsewhere in the literature: improvements in school attendance and enrollment as well as increases in health care utilization. Although many cash transfers are conditioned on nutritional behaviors for young children, the literature finds heterogeneous or modest impacts on nutrition (Gertler, 2004; 14 There is also a 10% increase in the supply of local mid-wives, but the increase is not sufficient to prevent a pricerise. 20

21 Behrman and Hoddinott, 2005), and typically not on anthropometric outcomes (Alatas et al., 2011; Fiszbein et al., 2009). In contrast, we find a direct and positive impact on child nutrition from the Pantawid program: severe stunting among 6-36 month old children was 8.5 percentage points lower in treated areas than in control ones. However, this gain was accompanied by a worsening of nutritional status among nonbeneficiary househ, particularly for children who are in vulnerable ages (the first 1000 days) when the program rolled out. Moreover, when program exposure at the village level (which we term saturation) is higher, the detrimental impacts on non-beneficiaries are larger. And second, prices for perishable protein-rich foods are higher when saturation is higher; prices are measured both through the survey respondent report of unit values as well as official price data at the provincial level. These effects are not simply short-run effects; the timeline between implementation and survey was 31 months on average. In theory, the general price effects from a cash transfer are magnified as the number of beneficiaries increases, as well as if the local market that supplies goods is not fully integrated in the wider economy thus allowing local demand and supply conditions to largely determine the price. This lack of integration can arise either in remote areas, with consequent high transport costs into and out of the local market, or for perishable goods that require cooling technologies to transport and store and hence also exhibit higher costs of trade over long distances. In these cases the elasticity of supply to a demand increase may be low, especially with respect to marginal changes in price. The food goods examined here are a combination of locally produced and largely perishable food goods along with packaged goods largely traded in national markets. It is these perishable food goods that appear to exhibit modest price increases after the introduction of the program, especially in the saturated villages that experienced the largest increases in non-beneficiary child stunting. Taken together, the findings suggest that anti-poverty cash transfer programs may have unanticipated effects operating through the price channel, especially for noncompensated (non-beneficiary) househ. These effects will manifest or be greater in local markets that have a high degree of program coverage and/or markets that are less integrated into the national economy. Besides the food price channel, additional channels that may contribute to the worsening of child anthropometric status include utilization spillovers in the formal health care system, and changes in the availability of young child caretakers. Little evidence was found for spillovers in the health system, although the analysis could not investigate possible quality of care changes. There is some indication that the presence of adults is lessened in the non-beneficiary househ in the treated saturated areas both in terms of a lower number of adults in the household and an increase in the hours worked by adult laborers but this effect is not 21

22 precisely estimated and does not appear to be quantitatively large. The analysis cannot completely rule out these complementary channels that may also explain the program spillover on stunting, partly due to the fact that important dimensions of these channels are unobserved. Yet if they arise they would operate alongside the price channel and may, to some extent, but a result of the village-level price changes brought on by the program. Given the unintended negative consequences for young children in non-beneficiary househ, and the fact that these consequences arise in poorer and more remote villages, the question of targeting rules comes to the forefront. The Pantawid program is targeted to the individual household on the basis of its proxy-means test score. However, for the subset of villages that are particularly poor and/or remote, a village based targeting scheme would presumably compensate all househ for any rise in local prices and thus avert increases in child stunting. However area based targeting, while averting spillovers, would also likely be more expensive. In order to provide an order of magnitude for the ratio of benefits to costs of extending the program coverage in barangays with high poverty rates to the entire barangay, we carry out an exercise relating (i) the discounted value of labor market returns to averting stunting at age 36 months to (ii) the discounted costs of the family transfer associated with adding one household, with one child, to the program roll. Note that the benefits in this exercise are narrowly defined to those associated with lifetime earnings. In order to estimate the impact of stunting on labor market returns, we use the parameter estimate from Hoddinott et al (2011) who find that hourly earnings among adults who were stunted at age 36 months are 0.58 times the hourly earnings of those who were not stunted, after controlling for a number of contextual factors. Based on the average daily adult wages reported in our sample (US$6.3), we assume that the annual earnings of an adult who was stunted as a child are 0.58 times those of an adult who was not stunted as a child in each year that they work (following the method sketched in Hoddinott et al 2013). Since we find that Pantawid increased the prevalence of stunting by 12 percentage points among non-beneficiary children, we further multiply this value by 0.12 to estimate only the value of the stunting differential that can be attributed to the program. Using these parameters, we estimate that the discounted lifetime benefits of the program s impact on stunting (manifested in lifetime earnings) equals the discounted program costs when (i) we assume that real wages will grow at a rate of 1.75 percent per year, which is close to the rate observed in 2012 and 2013, 15 and (ii) apply a discount rate of 5 percent. At any lower discount rate (holding projected real wage growth constant) the benefit/cost ratio is positive; at any higher projected real wage growth (holding the discount rate constant) the benefit/cost ratio is positive and 1.5 percent real wage growth reported for 2012 and 2013 respectively in ILO (2014), 22

23 To fix ideas, the above estimates assume an annual family transfer of US$132 for when the child is aged 1 to 14 (that is, the basic transfer amount); discounted back to age 0, this amounts to a total per-child value of the transfer (i.e. the cost) of US$1,636. Given the real wage growth estimate of 1.75% and a discount rate of 5%, and attributing 12% of the gap in earnings between adults who were stunted and those who were not stunted to the program, this is also equal to lifetime earnings detriment associated with the program. Of course an exercise such as this is sensitive to a number of assumptions. In this case in particular, the labor market penalty associated with having been stunted at age 36 months (drawn from Hoddinott et al. 2011) appears to be quite high. We estimate that the program would have a benefit/cost ratio of greater than 1 as long as hourly wages for those who were stunted are 0.75 or less than those who were not stunted. Further we have simplified the cost implications of switching from a household to a village targeting mechanism. Adding more househ to the beneficiary rolls would undoubtedly increase total administrative cost to some degree, yet at the same time the adoption of a village-based targeting rule may require far less household information to be collected and so would likely provide savings in this dimension. Further work needs to be done to estimate more comprehensively the lifetime benefits of averting stunting as well as the programmatic costs of different targeting mechanisms. However the initial estimate here suggests that a national program may wish to consider a hybrid targeting scheme for their anti-poverty programs when faced with the possibility of local market price spillovers to non-beneficiaries in poorer and more remote villages. For these villages, offering the program to every household may be more cost-effective. Other areas of the country that likely will not experience local price spillovers can continue with targeting the program only to the poorest househ. 23

24 References Alatas, Vivi, Nur Cahyadi, Elisabeth Ekasari, Sarah Harmoun, Budi Hidayat, Edgar Janz, Jon Jellema, Hendratno Tuhiman, and Matthew Wai-Poi Main Findings from the Impact Evaluation of Indonesia's Pilot Household Conditional Cash Transfer Program. World Bank. Alix-Garcia, J., McIntosh, C., Sims, K. R., and Welch, J. R The ecological footprint of poverty alleviation: evidence from Mexico's Oportunidades program. Review of Economics and Statistics, 95(2), Angelucci, Manuela and Giacomo Di Giorgi Indirect Effects of an Aid Program: How Do Cash Transfers Affect Ineligibles' Consumption? The American Economic Review. 99(1): Baird, Sarah, et al "Relative effectiveness of conditional and unconditional cash transfers for schooling outcomes in developing countries: a systematic review." Campbell Systematic Reviews 9.8. Baird, Sarah; Bohren, Aislinn, McIntosh, Craig and Özler, Berk Designing Experiments to Measure Spillover Effects. World Bank Policy Research Working Paper Series Washington, DC. World Bank.Beegle, Kathleen, Emanuela Galasso and Jessica Goldberg Direct and Indirect Effects of Malawi s Public Works Program on Food Security. World Bank Policy Research Working Paper Series Washington, DC. World Bank. Behrman, Jere R. and John Hoddinott "Programme evaluation with unobserved heterogeneity and selective implementation: The Mexican PROGRESA impact on child nutrition." Oxford bulletin of economics and statistics 67.4: Bobba, Matteo and Jeremie Gignoux Policy Evaluation in the Presence of Spatial Externalities: Reassessing the Progresa Program. PSE Working Papers Bobonis, G. J. and Finan, F Neighborhood peer effects in secondary school enrollment Decisions. Review of Economics and Statistics 91(4), Bourguignon, F., Ferreira, F. H. and Leite, P. G Conditional cash transfers, schooling, and child labor: Micro-simulating Brazil's Bolsa Escola program. The World Bank Economic Review, 17(2), Chaudhury, Nazmul, Jed Friedman and Junko Onishi "Philippines conditional cash transfer program impact evaluation 2012." World Bank. Contreras, Diana and Pushkar Maitra Health Spillover Effects of a Conditional Cash Transfer Program. No Monash University, Department of Economics. 24

25 Crost, Benjamin, Joseph Felter and Patrick Johnston "Aid under fire: Development projects and civil conflict." The American Economic Review104.6: Cunha, Jesse M., Giacomo De Giorgi and Seema Jayachandran The price effects of cash versus in-kind transfers. No. w National Bureau of Economic Research. De Hoop, Jacobus, Jed Friedman, Eeshani Kandpal and Furio Rosati Can a partial schooling subsidy increase child labor? Experimental evidence from the Philippines and Mexico. Working Paper. DFID Cash Transfers: Literature Review. Department for International Development. DSWD Pantawid Pamilyang Pilipino Program. The Government of The Philippines. Ferreira, Francisco H., Deon Filmer and Norbert Schady "Own and sibling effects of conditional cash transfer programs: Theory and evidence from Cambodia." World Bank Policy Research Working Paper Series Washington, DC. World Bank. Fiszbein, Ariel and Norbert Schady Conditional cash transfers: reducing present and future poverty. World Bank. Gertler, Paul Do conditional cash transfers improve child health? Evidence from PROGRESA's control randomized experiment." The American Economic Review, 94(2): Hanlon, Joseph, Armando Barrientos, and David Hulme Just give money to the poor: The development revolution from the global South. Kumarian Press. Hoddinott, John, John Maluccio, Jere R. Behrman, Reynaldo Martorell, Paul Melgar, Agnes R. Quisumbing, Manuel Ramirez-Zea, Aryeh D. Stein, Kathryn M. Yount The Consequences of Early Childhood Growth Failure over the Life Course. International Food Policy Research Institute Discussion Paper No Hoddinott, John, Harold Alderman, Jere R. Behrman, Lawrence Haddad and Susan Horton The economic rationale for investing in stunting reduction. Maternal and Child Nutrition 9(Suppl. 2): Labonne, Julien "The local electoral impacts of conditional cash transfers: Evidence from a field experiment." Journal of Development Economics 104: Lalive, R. and Cattaneo, M. A Social interactions and schooling decisions, The Review of Economics and Statistics 91(3), Macours, Karen, Norbert Schady, and Renos Vakis "Cash transfers, behavioral changes, and cognitive development in early childhood: Evidence from a randomized experiment." American Economic Journal: Applied Economics 4.2:

26 Nazara, Suahasil and Sri Kusumastuti Rahayu Program Kelyarga Harapan (PKH): Indonesian Conditional Cash Transfer Programme. International Policy Center for Inclusive Growth: Policy Brief No. 42. Philippine Statistics Authority Price Indices. The Government of The Philippines. Web. 10 November Puentes, Esteban, Fan Wang, Jere R. Behrman, Flavio Cunha, John Hoddinott, John A. Maluccio, Linda S. Adair, Reynaldo Martorell, and Aryeh D. Stein "Early Life Height and Weight Production Functions with Endogenous Energy and Protein Inputs." Rutstein, SO and G Rojas. Guide to DHS statistics. Calverton, MD: ORC Macro, MEASURE DHS+; 2006 [cited 2016 March 8]. Available from: df Saavedra, J. E. and Garcia, S Impacts of conditional cash transfer programs on educational outcomes in developing countries: a meta-analysis.rand Labor and Population Working Paper Series, WR Teixeira, Clarissa Gondim, et al Externality and behavioural change effects of a nonrandomised CCT programme: Heterogeneous impact on the demand for health and education. No. 82. Working Paper, International Policy Centre for Inclusive Growth. Triyana, M Do Health Care Providers Respond to Demand-Side Incentives? Evidence from Indonesia. Working Paper. World Bank A Model from Mexico for the World. Web. November 15, WHO Multicentre Growth Reference Study Group. WHO Child Growth Standards: Length/height-for-age, weight-for-age, weight-for-length, weight-for-height and body mass index-for-age: Methods and development. Geneva: World Health Organization Available from: Accessed May 6,

27 Figures and Tables Change in Province-level Saturation Rates Figure 1: Expansion in the Coverage of Pantawid between 2008 and 2014, by Province Figure 2: Within-Village Variation in Program Saturation 27

28 Figure 3: Stunting Prevalence in Children Exposed to Pantawid in First 1000 Days Figure 4: Stunting Rates by Above and Below Median Exposure to Pantawid 28

29 Figure 5: Difference in Stunting Rates (Treatment Control) among Non-Beneficiaries, by Quartiles of Pantawid Saturation (or Exposure ) 29